The cyclical-replenishment permit schedules presented in Sect.4approximate real- world markets more poorly than random-replenishment permit schedules. In real markets, demand and supply does not arrive in neat price-ordered cycles. For that reason, where results from cyclical markets (presented in Sect.5.1) show a significant effect that is not also present in random markets, we interpret it as an indication that introducing artificial constraints into experimental markets for ease of analysis runs the risk of also introducing artefacts that, because they are statistically significant, can be misleading.
The following relationships were all observed to be statistically significant in cyclical markets and not statistically significant in random markets, providing further support for the argument forrealismin artificial-market experiment design, previously advanced at length in [19]:
1. Cyclical-replenishment markets have significantly greaterα in the first period of trade (see Sect.5.1.2). This is a direct consequence of cyclical-replenishment allocating orders in a monotonically decreasing sequence from most profitable to least profitable. As such, the first orders allocated into the market have limit prices far from equilibrium. Since the market is empty, there is no mechanism for price discovery other than trial-and-error exploration, leading to largeα. In random-replenishment markets, the initial orders entering the market are drawn at random from the demand and supply schedules. This leads to lower bounds on limit prices and hence lowerα. Subsequently, price discovery is led by the order book, resulting in lowerαthat is statistically similar in both cyclical and random markets.
64 J. Cartlidge and D. Cliff 2. In cyclical-replenishment markets, the efficiency of AA-0.1 robots is signifi- cantly higher than the efficiency of the other robot types (see Sect.5.1.3). While there is some evidence of an inverse relationship between robot sleep time and robot efficiency across all markets, we infer that this difference is an artefact of cyclical replenishment until further experimental trials can confirm otherwise.
3. In cyclical-replenishment markets, profit dispersion is significantly higher for agents, humans, and the market as a whole (see Sect.5.1.5). Since lower profit dispersion is a desirable property of a market, this suggests that the relatively high profit dispersion observed in previous cyclical-replenishment experiments [11,19] is an artefact of the experimental design.
7 Conclusion
We have presented a series of laboratory experiments between agent traders and human traders in a controlled financial market. Results demonstrate that, despite the simplicity of the market, when agents act on super-human timescales—i.e., when the sleep-wake cycle of agents is 0.1 s—the market starts to fragment, such that agents are more likely to trade with agents, and humans are more likely to trade with humans. In contrast, when agents act on human timescales—i.e., when the sleep- wake cycle of agents is 1 s, or above—the markets are well mixed, with agents and humans equally likely to trade between themselves and between each other. This transition to a fragmented market from a mixed market intriguingly appears to be linked to market inefficiency, such that below the threshold of human reaction times (i.e., at 0.1 s timescale) any idiosyncratic agent behaviours can adversely perturb the market; whereas above the threshold (i.e., at timescales of 1 s and above) human interactions help to dampen market perturbations, ensuring better equilibration and efficiency.
This behaviour has parallels with the real financial markets, and in particular, we present this as tantalising evidence for the robot phase transition (RPT), discovered by Johnson et al. [35,36]. In Johnson et al.’s words, “a remarkable new study by Cliff and Cartlidge provides some additional support for our findings. In controlled lab experiments, they found when machines operate on similar timescales to humans (longer than 1 s), the ‘lab market’ exhibited an efficient phase (c.f. few extreme price-change events in our case). By contrast, when machines operated on a timescale faster than the human response time (100 ms) then the market exhibited an inefficient phase (c.f. many extreme price-change events in our case)” [36].
In the final quarter of 2016, a new exchange node containing the first ever intentional delay was introduced in the USA. To achieve a delay of 350às in signal transmission, the exchange embedded a 38-mile coil of fibre optic cable.
The desired intention is to “level out highly asymmetric advantages available to faster participants” in the market [34]. However, the impact this might have at the system level is unknown. To address this, Johnson declares that more academic studies need to focus on subsecond resolution data, and he identifies the work we
Modelling Financial Markets Using Human–Agent Experiments 65 have reported here as one of the few exceptions in the literature that attempts to understand subsecond behaviours [34].
This work is presented as a demonstration of the utility of using experimental human–agent laboratory controlled markets: (a) to better understand real-world complex financial markets; and (b) to test novel market policies and structures before implementing them in the real world. We hope that we are able to encourage the wider scientific community to pursue more research endeavour using this methodology.
Future Work
For results presented here, we used De Luca’s OpEx experimental trading software, running on the Lab-in-a-box hardware, a self-contained wired-LAN containing networked exchange server, netbooks for human participants, and an administrator’s laptop. This platform is ideally suited for controlled real-time trading experiments, but is designed for relatively small-scale, synchronous markets where participants are physically co-located. If experiments are to be scaled up, to run for much longer periods and to support large-scale human participation, an alternative platform architecture is required. To this end, development began on ExPo—theExchange Portal—in 2011. ExPo has a Web service architecture, with humans participating via interaction through a Web browser (see [50]). This enables users to connect to the exchange via the Internet, and participate remotely. Immediately, ExPo negates the requirement for specific hardware, and enables long-term and many-participant experimentation, with users able to leave and return to a market via individual account log-in. Currently, an updated version—ExPo2—is under development at UNNC, in collaboration with Paul Dempster. As with OpEx and ExPo, ExPo2 will be released open-source to encourage replication studies and engagement in the wider scientific community.
In [8] a detailed proposal for future research studies is presented. In particular, future work will concentrate on relaxing some experimental constraints, such as enabling agents to trade on their own account, independent of permit schedules.
This relaxation—effectively changing the function of agents from an agency trader (or “broker”) design to a proprietary “prop” trader design—should enable the emergence of more realistic dynamics, such as Johnson et al.’s UEE price swing fractures. If we are able to reproduce these dynamics in the lab, this will provide compelling evidence for the RPT. Further, market structures and regulatory mechanisms such as financial circuit breakers, intentional network delays, and periodic (rather than real-time) order matching at the exchange will be tested to understand the impact these have on market dynamics. In addition, preliminary studies to monitor human emotional responses to market shocks, using EEG brain data, are underway. Hopefully these studies can help us better understand how emotional reactions can exacerbate market swings, and how regulatory mechanisms, or trading interface designs, can be used to dampen such adverse dynamics.
Acknowledgements The experimental research presented in this chapter was conducted in 2011–
2012 at the University of Bristol, UK, in collaboration with colleagues Marco De Luca (the
66 J. Cartlidge and D. Cliff
developer of OpEx) and Charlotte Szostek. They both deserve a special thanks. Thanks also to all the undergraduate students and summer interns (now graduated) that helped support related work, in particular Steve Stotter and Tomas Gražys for work on the original ExPo platform.
Finally, thanks to Paul Dempster and the summer interns at UNNC for work on developing the ExPo2 platform, and the pilot studies run during July 2016. Financial support for the studies at Bristol was provided by EPSRC grants EP/H042644/1 and EP/F001096/1, and funding from the UK Government Office for Science (Go-Science) Foresight Project onThe Future of Computer Trading in Financial Markets.Financial support for ExPo2 development at UNNC was provided by FoSE summer internship funding.
References
1. Angel, J., Harris, L., & Spratt, C. (2010). Equity trading in the 21st century. Working Paper FBE-09-10, Marshall School of Business, University of Southern California, February 2010.
Available via SSRNhttps://ssrn.com/abstract=1584026Accessed 22.03.2017 2. Arthur, W. B. (2014).Complexity and the economy. Oxford: Oxford University Press.
3. Battiston, S., Farmer, J. D., Flache, A., Garlaschelli, D., Haldane, A. G., Heesterbeek, H., et al.
(2016). Complexity theory and financial regulation: Economic policy needs interdisciplinary network analysis and behavioral modeling.Science, 351(6275), 818–819
4. Baxter, G., & Cartlidge, J. (2013). Flying by the seat of their pants: What can high frequency trading learn from aviation? In G. Brat, E. Garcia, A. Moccia, P. Palanque, A. Pasquini, F.
J. Saez, & M. Winckler (Eds.),Proceedings of 3rd International Conference on Applied and Theory of Automation in Command and Control System (ATACCS), Naples (pp. 64–73). New York: ACM/IRIT Press, May 2013.
5. Berger, S. (Ed.), (2009).The foundations of non-equilibrium economics. New York: Routledge.
6. Bisias, D., Flood, M., Lo, A. W., & Valavanis, S. (2012). A survey of systemic risk analytics.
Annual Review of Financial Economics, 4, 255–296.
7. Bouchaud, J. P. (2008). Economics needs a scientific revolution.Nature, 455(7217), 1181.
8. Cartlidge, J. (2016). Towards adaptive ex ante circuit breakers in financial markets using human-algorithmic market studies. In Proceedings of 28th International Conference on Artificial Intelligence(ICAI), Las Vegas (pp. 77–80). CSREA Press, Athens, GA, USA. July 2016.
9. Cartlidge, J., & Cliff, D. (2012). Exploring the ‘robot phase transition’ in experimental human-algorithmic markets. In Future of computer trading. Government Office for Science, London, UK (October 2012) DR25. Available via GOV.UKhttps://www.gov.uk/government/
publications/computer-trading-robot-phase-transition-in-experimental-human-algorithmic- marketsAccessed 22.03.2017.
10. Cartlidge, J., & Cliff, D. (2013). Evidencing the robot phase transition in human-agent experimental financial markets. In J. Filipe & A. Fred (Eds.),Proceedings of 5th International Conference on Agents and Artificial Intelligence (ICAART), Barcelona (Vol. 1, pp. 345–352).
Setubal: SciTePress, February 2013.
11. Cartlidge, J., De Luca, M., Szostek, C., & Cliff, D. (2012). Too fast too furious: Faster financial- market trading agents can give less efficient markets. In J. Filipe & A. Fred (Eds.),Proceedings of 4th International Conference on Agents and Artificial Intelligent (ICAART), Vilamoura (Vol. 2, pp. 126–135). Setubal: SciTePress, February 2012.
12. Chen, S. H., & Du, Y. R. (2015). Granularity in economic decision making: An interdisci- plinary review. In W. Pedrycz & S. M. Chen (Eds.),Granular computing and decision-making:
Interactive and iterative approaches(pp. 47–72). Berlin: Springer (2015)
13. Cliff, D., & Bruten, J. (1997).Minimal-Intelligence Agents for Bargaining Behaviours in Market-Based Environments. Technical Report HPL-97-91, Hewlett-Packard Labs., Bristol, August 1997.
Modelling Financial Markets Using Human–Agent Experiments 67
14. Cliff, D., & Northrop, L. (2017). The global financial markets: An ultra-large-scale systems perspective. In: Future of computer trading. Government Office for Science, London, UK (September 2011) DR4. Available via GOV.UKhttps://www.gov.uk/government/publications/
computer-trading-global-financial-marketsAccessed 22.03.2017
15. Das, R., Hanson, J., Kephart, J., & Tesauro, G. (2001) Agent-human interactions in the continuous double auction. In Nebel, B. (Ed.),Proceedings of 17th International Conference on Artificial Intelligence (IJCAI), Seattle (pp. 1169–1176). San Francisco: Morgan Kaufmann, August 2001
16. De Luca, M. (2015). Why robots failed: Demonstrating the superiority of multiple-order trading agents in experimental human-agent financial markets. In S. Loiseau, J. Filipe, B.
Duval, & J. van den Herik, (Eds.),Proceedings of 7th International Conference on Agents and Artificial Intelligence (ICAART), Lisbon (Vol. 1, pp. 44–53). Setubal: SciTePress, January 2015.
17. De Luca, M., & Cliff, D. (2011). Agent-human interactions in the continuous double auction, redux: Using the OpEx lab-in-a-box to explore ZIP and GDX. In J. Filipe, & A. Fred (Eds.), Proceedings of 3rd International Conference on Agents and Artificial Intelligents (ICAART) (Vol. 2, pp. 351–358) Setubal: SciTePress, January 2011.
18. De Luca, M., & Cliff, D. (2011). Human-agent auction interactions: Adaptive-aggressive agents dominate. In Walsh, T. (Ed.),Proceedings of 22nd International Joint Conference on Artificial Intelligence (IJCAI)(pp. 178–185). Menlo Park: AAAI Press, July 2011.
19. De Luca, M., Szostek, C., Cartlidge, J., & Cliff, D. (2011). Studies of interactions between human traders and algorithmic trading systems. In: Future of Computer Trading. Government Office for Science, London, September 2011, DR13. Available via GOV.UK https://www.
gov.uk/government/publications/computer-trading-interactions-between-human-traders-and- algorithmic-trading-systemsAccessed 22.03.17.
20. Easley, D., & Kleinberg, J. (2010).Networks, crowds, and markets: Reasoning about a highly connected world. Cambridge: Cambridge University Press
21. Easley, D., Lopez de Prado, M., & O’Hara, M. (Winter 2011). The microstructure of the ‘flash crash’: Flow toxicity, liquidity crashes and the probability of informed trading.Journal of Portfolio Management, 37(2), 118–128
22. Farmer, J. D., & Foley, D. (2009). The economy needs agent-based modelling. Nature, 460(7256), 685–686
23. Farmer, J. D., & Skouras, S. (2011). An ecological perspective on the future of computer trading. In: Future of Computer Trading. Government Office for Science, London, September 2011, DR6. Available via GOV.UK https://www.gov.uk/government/publications/computer- trading-an-ecological-perspectiveAccessed 22.03.2017.
24. Feltovich, N. (2003). Nonparametric tests of differences in medians: Comparison of the Wilcoxon-Mann-Whitney and Robust Rank-Order tests.Experimental Economics, 6, 273–297.
25. Foresight. (2012). The Future of Computer Trading in Financial Markets. Final project report, The Government Office for Science, London, UK (October 2012).
Available via GOV.UK http://www.cftc.gov/idc/groups/public/@aboutcftc/documents/file/
tacfuturecomputertrading1012.pdfAccessed 22.03.17
26. Giles, J. (2012). Stock trading ‘fractures’ may warn of next crash. New Scientist (2852) (February 2012). Available Online: http://www.newscientist.com/article/mg21328525.700- stock-trading-fractures-may-warn-of-next-crash.htmlAccessed 22.03.17.
27. Gjerstad, S., & Dickhaut, J. (1998). Price formation in double auctions.Games and Economic Behavior, 22(1), 1–29
28. Gode, D., & Sunder, S. (1993). Allocative efficiency of markets with zero-intelligence traders:
Markets as a partial substitute for individual rationality.Journal of Political Economy, 101(1), 119–137.
29. Gomber, P., Arndt, B., Lutat, M., & Uhle, T. (2011).High Frequency Trading. Technical report, Goethe Universitọt, Frankfurt Am Main (2011). Commissioned by Deutsche Bửrse Group.
68 J. Cartlidge and D. Cliff
30. Grossklags, J., & Schmidt, C. (2003). Artificial software agents on thin double auction markets:
A human trader experiment. In J. Liu, B. Faltings, N. Zhong, R. Lu, & T. Nishida (Eds.), Proceedings of IEEE/WIC Conference on Intelligent Agent and Technology (IAT), Halifax (pp. 400–407). New York: IEEE Press.
31. Grossklags, J., & Schmidt, C. (2006). Software agents and market (in)efficiency: A human trader experiment.IEEE Transactions on Systems, Man and Cybernetics, Part C (Applications and Review) 36(1), 56–67.
32. Holt, C. A., & Roth, A. E. (2004). The Nash equilibrium: A perspective.Proceedings of the National Academy of Sciences of the United States of America, 101(12), 3999–4002
33. Huber, J., Shubik, M., & Sunder, S. (2010). Three minimal market institutions with human and algorithmic agents: Theory and experimental evidence.Games and Economic Behavior, 70(2), 403–424
34. Johnson, N. (2017). To slow or not? Challenges in subsecond networks.Science, 355(6327), 801–802.
35. Johnson, N., Zhao, G., Hunsader, E., Meng, J., Ravindar, A., Carran, S., et al. (2012).
Financial Black Swans Driven by Ultrafast Machine Ecology. Working paper published on arXiv repository, Feb 2012.
36. Johnson, N., Zhao, G., Hunsader, E., Qi, H., Johnson, N., Meng, J., et al. (2013). Abrupt rise of new machine ecology beyond human response time.Scientific Reports, 3(2627), 1–7 (2013) 37. Joint CFTC-SEC Advisory Committee on Emerging Regulatory Issues. (2010). Findings Regarding the Market Events of May 6, 2010. Report, CTFC-SEC, Washington, DC, Septem- ber 2010. Available via SEChttps://www.sec.gov/news/studies/2010/marketevents-report.pdf Accessed 22.03.2017.
38. Kahneman, D. (2011).Thinking, fast and slow. New York, NY: Farrar, Straus and Giroux.
39. Keim, B. (2012). Nanosecond trading could make markets go haywire. Wired (February 2012).
Available Online:http://www.wired.com/wiredscience/2012/02/high-speed-tradingAccessed 22.03.2017.
40. Leinweber, D. (2009).Nerds on wall street. New York: Wiley.
41. May, R. M., Levin, S. A., & Sugihara, G. (2008) Complex systems: Ecology for bankers.
Nature, 451, 893–895
42. Nelson, R. H. (2001). Economics as religion: From Samuelson to Chicago and beyond.
University Park, PA: Penn State University Press.
43. Nelson, R. R., & Winter, S. G. (1982).An evolutionary theory of economic change. Harvard:
Harvard University Press.
44. Perez, E. (2011).The speed traders. New York: McGraw-Hill.
45. Price, M. (2012). New reports highlight HFT research divide. Financial News (February 2012). Available Online: https://www.fnlondon.com/articles/hft-reports-highlight-research- divide-cornell-20120221Accessed 22.03.2017.
46. Schweitzer, F., Fagiolo, G., Sornette, D., Vega-Redondo, F., & White, D. R. (2009). Economic networks: what do we know and what do we need to know?Advances in Complex Systems, 12(04n05), 407–422
47. Smith, V. (1962). An experimental study of comparative market behavior.Journal of Political Economy, 70, 111–137
48. Smith, V. (2006).Papers in experimental economics. Cambridge: Cambridge University Press.
49. Stotter, S., Cartlidge, J., & Cliff, D. (2013). Exploring assignment-adaptive (ASAD) trading agents in financial market experiments. In J. Filipe & A. Fred (Eds.), Proceedings of 5th International Conference on Agents and Artificial Intelligence (ICAART), Barcelona. Setubal:
SciTePress, February 2013.
50. Stotter, S., Cartlidge, J., & Cliff, D. (2014). Behavioural investigations of financial trading agents using Exchange Portal (ExPo). In N. T. Nguyen, R. Kowalczyk, A. Fred, & F. Joaquim (Eds.),Transactions on computational collective intelligence XVII. Lecture notes in computer science (Vol. 8790, pp. 22–45). Berlin: Springer.
Modelling Financial Markets Using Human–Agent Experiments 69
51. Tesauro, G., & Bredin, J. (2002). Strategic sequential bidding in auctions using dynamic programming. In C. Castelfranchi & W. L. Johnson (Eds.),Proceedings of 1st International Conference on Autonomous Agents and Multiagent Systems (AAMAS), Bologna (pp. 591–598).
New York: ACM.
52. Tesauro, G., & Das, R. (2001). High-performance bidding agents for the continuous double auction. InProceedings of the ACM Conference on Electronic Commerce (EC), Tampa, FL (pp. 206–209), October 2001.
53. Treanor, J. (2017). Pound’s flash crash ‘was amplified by inexperienced traders’. The Guardian, January 2017. Available Online https://www.theguardian.com/business/2017/jan/13/pound- flash-crash-traders-sterling-dollarAccessed 22.03.2017.
54. Vytelingum, P. (2006).The Structure and Behaviour of the Continuous Double Auction. PhD thesis, School of Electronics and Computer Science, University of Southampton.
55. Vytelingum, P., Cliff, D., & Jennings, N. (2008). Strategic bidding in continuous double auctions.Artificial Intelligence, 172, 1700–1729
56. Widrow, B., & Hoff, M. E. (1960). Adaptive switching circuits. Institute of Radio Engineers, Western Electronic Show and Convention, Convention Record, Part 4 (pp. 96–104).
57. Zhang, S. S. (2013).High Frequency Trading in Financial Markets. PhD thesis, Karlsruher Institut für Technologie (KIT).
Does High-Frequency Trading Matter?
Chia-Hsuan Yeh and Chun-Yi Yang
Abstract Over the past few decades, financial markets have undergone remark- able reforms as a result of developments in computer technology and changing regulations, which have dramatically altered the structures and the properties of financial markets. The advances in technology have largely increased the speed of communication and trading. This has given birth to the development of algorithmic trading (AT) and high-frequency trading (HFT). The proliferation of AT and HFT has raised many issues regarding their impacts on the market. This paper proposes a framework characterized by an agent-based artificial stock market where market phenomena result from the interaction between many heterogeneous non-HFTs and HFTs. In comparison with the existing literature on the agent-based modeling of HFT, the traders in our model adopt a genetic programming (GP) learning algorithm.
Since they are more adaptive and heuristic, they can form quite diverse trading strategies, rather than zero-intelligence strategies or pre-specified fundamentalist or chartist strategies. Based on this framework, this paper examines the effects of HFT on price discovery, market stability, volume, and allocative efficiency loss.
Keywords High-frequency trading ã Agent-based modeling ã Artificial stock market ã Continuous double action ã Genetic programming
C.-H. Yeh ()
Department of Information Management, Yuan Ze University, Taoyuan, Chungli, Taiwan e-mail:imcyeh@saturn.yzu.edu.tw
C.-Y. Yang
Department of Computational and Data Sciences, College of Science, Krasnow Institute for Advanced Study, George Mason University, Fairfax, VA, USA
© Springer Nature Switzerland AG 2018
S.-H. Chen et al. (eds.),Complex Systems Modeling and Simulation in Economics and Finance, Springer Proceedings in Complexity, https://doi.org/10.1007/978-3-319-99624-0_4
71